Traditional transition region extraction methods depend much on the clip limits Llow and Lhigh. In which methods Llow and Lhigh can not often be obtained correctly from real images, which will result in incorrect extraction of transition region and finally bad quality of segmentation. A novel gradient threshold-based transition region extraction method (GT-TREM) is presented. Transition regions can steadily be extracted by GT-TREM. Experimental results demonstrated the robustness and effectiveness of the algorithm.